Anthropic's Revenue Grew Sevenfold in Seven Months, and Just Passed OpenAI's

Some weeks in AI are about a single dramatic breakthrough. This one was about money, permits, and report cards. Anthropic quietly became bigger than its biggest rival by revenue. Nvidia found a way to buy into a startup’s future without buying the startup. Nevada let thousands of robotaxis loose on the Las Vegas Strip. And a new nonprofit graded the industry’s safety habits, with results nobody should be proud of. In India, a government ministry and the country’s biggest payments app found a new use for transaction data, and a solar project in Gujarat quietly finished another chunk of one of the country’s largest hybrid power plants.

The AI lab that quietly got bigger than OpenAI

Anthropic’s annualized revenue run rate, a measure of what the company would collect over a year at its current pace, climbed from about $9 billion at the end of 2025 to more than $65 billion by the end of July. That is more than sevenfold growth in seven months, and it now puts Anthropic ahead of OpenAI, whose own run rate sits at roughly $40 billion.

The pace kept accelerating even within that stretch. Anthropic’s run rate was $47 billion in May, meaning it added close to $18 billion more in just two months. Its actual second quarter revenue, the money collected between April and June rather than a rate projected forward, came in above $11.5 billion, more than 14 times what the company brought in over the same quarter a year earlier.

Anthropic confidentially filed a draft registration statement with the Securities and Exchange Commission on June 1, which gives it the option to go public without committing to a date. Multiple reports this week say the company could make that filing public as soon as the end of August, with a listing possible this fall at a valuation north of $2 trillion. That would put it in the same league as SpaceX, which raised $75 billion in its own record breaking IPO in June and closed its first day of trading worth about $2.2 trillion.

Nevada just handed out 8,000 robotaxi permits in one vote

The Nevada Transportation Authority voted unanimously this week to let Tesla, Uber and Waymo run commercial robotaxi services across Clark County, home to Las Vegas. Between the three companies, the permits allow for up to 8,000 self driving cars on the road over the next year: Tesla up to 5,000, Waymo up to 1,000, and Uber another 1,000 through partnerships with Motional and Zoox.

The scale of the shift is easy to miss without the backstory. Only days earlier, Nevada had Tesla capped at 10 robotaxis on a short stretch of the Strip, with a 45 mile per hour speed limit and airport trips off the table. The new permit raises that ceiling roughly 500 times over in a single vote. Even Tesla’s own robotaxi engineer has tempered expectations, saying he would be satisfied reaching 2,500 vehicles rather than the full 5,000 within the year.

Nvidia pays $6 billion to rent a rival’s brain, not buy it

Nvidia is paying $6 billion to license the technology behind Poolside, a startup that builds open weight coding models, without buying the company outright. The deal covers Poolside’s “Model Factory,” the system behind its Laguna family of models, and Nvidia will make job offers to the 109 Poolside staff who worked on that project. Poolside’s three co-founders stay in place and the company keeps operating independently.

Separately, Nvidia is putting $1 billion directly into Poolside through a growth equity investment, valuing the startup at $13 billion afterward. Paying for access and talent while leaving a company technically independent has become one of Nvidia’s favored ways to plant a flag inside AI labs it doesn’t want to fully own.

A safety report card grades every major AI lab, and nobody passes

GuideLight AI Standards, a new nonprofit founded by two former OpenAI safety and policy staffers, published its first assessment of how well the biggest AI labs actually control their own systems. It checked six practices: logging what an AI system is doing, measuring whether that monitoring works, requiring a check before risky actions go through, being able to halt a system automatically if something goes wrong, letting outside reviewers assess the work, and having an actual plan for containing a model that misbehaves.

Anthropic and OpenAI came out on top, both with a C+. Google scored a D+, xAI a D-, and Meta an F. GuideLight found that no company scored better than “substantial partial implementation” on any single practice, and most scores landed at “limited” or lower. Even at the labs trying hardest, the industry’s safety habits are closer to a rough draft than a finished system.

India’s biggest payments app just became an infrastructure planning tool

India’s Ministry of Electronics and Information Technology signed an agreement this week with PhonePe to feed data from the company’s new PulsePro platform into PM GatiShakti, the government’s national infrastructure planning portal. PulsePro turns anonymized transaction data from PhonePe’s user base, more than 700 million registered users and 50 million merchants spanning nearly every postal code in the country, into local, near real time signals about where economic activity is actually happening on the ground.

The goal is to give planners a live read on commerce and construction rather than waiting on periodic surveys when deciding where the next road, warehouse or power line should go. BISAG-N, the government’s geospatial applications institute, will help manage the integration.

Gujarat switches on more of one of India’s biggest solar plants

KPI Green Energy energized another 130 megawatts of solar capacity this week at its hybrid wind and solar plant in Bharuch, Gujarat, under a long term power purchase agreement with the state’s electricity utility. That brings the project, aimed at a total of 370 megawatts of AC capacity, to roughly 270 megawatts energized so far, about three quarters of the way to completion.

It’s one incremental step inside a much larger buildout, not a ribbon cutting ceremony. But it’s real megawatts feeding a real grid.

None of these stories close a chapter. Anthropic hasn’t filed its IPO yet, Nevada’s robotaxis haven’t all hit the road, and a C+ safety grade is still a mark nobody should celebrate. Together they’re a decent snapshot of where AI actually stands right now: less a single dramatic leap than a dozen labs, regulators and grid operators quietly deciding how much of the future to hand it, one contract and one vote at a time.